PaperlayerView exact revision on arXiv

Source-linked AI summary

Trademark Search, Artificial Intelligence and the Role of the Private Sector

Sonia Katyal, Aniket Kesari

arXiv:2601.17072v1cs.CYcs.AI

TL;DR

Trademark scholarship has largely emphasized consumer-side search while giving less attention to AI’s role in trademark selection and applicant-side costs. This paper combines a supply-side framework with experiments comparing private search engines and TESS, finding stronger recall for private engines but also possible new search costs. It concludes that AI’s effects on trademark registration and quality require greater attention to trademark applicants and the registration ecosystem.

  • Problem

    Existing trademark research focuses mainly on consumer search and gives limited attention to AI’s role in trademark selection and the substantial search costs faced by applicants.

  • Method

    The paper develops an applicant-centered framework and experimentally compares private AI-powered trademark search engines with TESS using exact, phonetic, precision, and recall measures.

  • Results

    Every private search engine achieves higher recall than TESS, while many improve on precision; search-engine behavior varies with result limits.

  • Takeaways & Limitations

    Trademark applicants should be analyzed as consumers within the trademark-selection ecosystem, with greater attention to trademark supply, registration, and quality.

Abstract

from arXiv · show

Almost every industry today confronts the potential role of artificial intelligence and machine learning in its future. While many studies examine AI in consumer marketing, less attention addresses AI's role in creating and selecting trademarks that are distinctive, recognizable, and meaningful to consumers. Traditional economic approaches to trademarks focus almost exclusively on consumer-based, demand-side considerations regarding search. However, these approaches are incomplete because they fail to account for substantial costs faced not just by consumers, but by trademark applicants as well. Given AI's rapidly increasing role in trademark search and similarity analysis, lawyers and scholars should understand its dramatic implications. This paper proposes that AI should interest anyone studying trademarks and their role in economic decision-making. We examine how machine learning techniques will transform the application and interpretation of foundational trademark doctrines, producing significant implications for the trademark ecosystem. We run empirical experiments regarding trademark search to assess the efficacy of various trademark search engines, many of which employ machine learning methods. Through comparative analysis, we evaluate how these AI-powered tools function in practice. In an age where artificial intelligence increasingly governs trademark selection, the classic division between consumers and trademark owners deserves an updated, supply-side framework. This insight has transformative potential for encouraging both innovation and efficiency in trademark law and practice.

A. SEARCH, EXPERIENCE, AND CREDENCE ATTRIBUTES IN CONSUMER

The supplied passage is a table-of-contents entry for a section on search, experience, and credence attributes in consumer decision-making.

  • The section is titled “Search, Experience, and Credence Attributes in Consumer Decision-Making.”
  • The entry places this discussion within the paper’s broader treatment of trademark decision-making.
  • The passage provides no substantive findings or analysis from the section.

II. SEARCH COSTS IN TRADEMARK REGISTRATION: A VIEW FROM A TRADEMARK APPLICANT .............................................. 514

The supplied passages identify a section on applicant-centered search costs and distinguish private-sector search engines from public search resources.

  • The section addresses search costs in trademark registration from a trademark applicant’s perspective.
  • The paper separately lists supplementing trademark search in the private sector as a topic.
  • It also lists artificial intelligence and intellectual-property administration as a related topic.
  • The contents identify a comparison of trademark search engines, including public and private search engines.
  • The listed private-search subsection includes Corsearch and Markify.

A. OUTCOMES AND IMPLICATIONS .............................................................. 573 B. FRAMING TRADEMARK REGISTRATION AS AN ADVERSARIAL

The paper presents AI as reshaping trademark research, search, registration, and economic analysis by adding an applicant-side, supply-oriented perspective. Its empirical comparison suggests that private AI-powered search engines can outperform TESS on recall, while AI may also create new search costs.

  • Framing and motivation: The paper addresses limited scholarship on AI’s role in trademarks despite broader research on AI in consumer marketing and other IP fields.
  • Framing and motivation: The authors study private trademark search engines and argue that AI will transform trademark creation, comparison, protection, and foundational legal doctrines.
  • Empirical study: The experiments compare AI-powered search engines with TESS for potential conflicts under Section 2(d) of the Trademark Act.
  • Implications: The paper reports that AI can reduce existing search costs but may introduce new search costs into the trademark ecosystem.
  • Implications: The authors propose treating trademark applicants as consumers of trademarks and shifting attention toward trademark supply and quality.

I. SEARCH COSTS IN TRADEMARK LAW: A VIEW FROM THE CONSUMER

This section develops the consumer-centered economics of trademark search and then broadens it toward applicant-side and AI-mediated search costs. It also identifies subjectivity and context dependence as boundaries on automated trademark analysis.

  • Consumer search costs: The paper situates trademark economics within Stigler’s framework for understanding information and consumer decision-making.
  • Consumer search costs: Traditional trademark analysis emphasizes how marks reduce consumer search costs by identifying products and their attributes.
  • Applicant-side search costs: The paper argues that conventional consumer-focused accounts overlook information markets affecting trademark supply and enforcement.
  • AI and search: The analysis distinguishes government AI assistance, government-provided tools for applicants, and private AI-driven tools.
  • AI and search: AI may assist trademark selection by analyzing visual, phonetic, semantic, classification, and other mark attributes.
  • Limits of automation: Subjective doctrines and context-specific judgments make trademark similarity and distinctiveness difficult for machines to capture fully.

1. Public Search Engines

Public and private trademark search engines differ in data sources, search features, and automation. TESS draws directly on the USPTO dataset, while private tools add services and algorithmic approaches intended to broaden or streamline searching.

  • Public Search Engines: TESS searches registered and applied-for marks but does not automatically flag conflicts, so users are advised to consult attorneys or search firms.Its search options vary in inclusiveness, from basic matching to more advanced use of design-mark codes and other information.
  • Public Search Engines: TESS draws directly on the authoritative USPTO trademark case-files dataset, which contains information about over eight million trademarks.
  • Private Search Engines: Corsearch’s screening product combines search results with visualization and document tools, while its phonetic engine covers phonetic, spelling, and plural variations.
  • Private Search Engines: Markify focuses on trademark searches and brand management, using a statistical similarity algorithm to help users prioritize results.
  • Private Search Engines: Trademarkia offers knockout and comprehensive searches; the study focuses on knockout searches because they appear algorithmic and do not involve human input.
  • Private Search Engines: TrademarkNow encodes legal rules and trademark-law intuitions into an AI model that uses domain models and machine-learning techniques to produce relevant results.

3. Our Methodology

The study evaluates trademark search engines by running standardized searches for marks already identified as confusingly similar and comparing the returned results. Its reproducible, programmatic pipeline supports scale and transparency, but the researchers cannot observe registrants’ actual searches or verify whether tested marks appeared in training data.

  • Our Methodology: The study compares how well trademark search engines identify potential conflicts under Section 2(d) of the Trademark Act.
  • Our Methodology: The end-to-end pipeline is programmatic and reproducible, reducing subjective judgment while enabling searches at scale.
  • Our Methodology: Researchers developed conflicted-mark lists, searched each term across all engines, saved the results, repeated the procedure, and analyzed precision, recall, and other metrics.
  • Generating Conflicted Trademarks: The researchers used recent 2(d) rejections because they reflect actual confusing-similarity decisions better than exact matches or artificially altered marks.
  • Limitations: The study cannot observe registrants’ actual searches or determine whether tested marks appeared in the search engines’ training data.
  • Our Methodology: Automated scripts also support repeated tests of search configurations, including translation, dead-mark, and match-type settings.

6. Exploratory Data Analysis

The exploratory analysis compares trademark search engines using result counts, exact matches, phonetic matches, and noise reduction. AI-powered private engines generally improve on TESS, but they optimize for different search objectives.

  • Comparison framework: TESS is the baseline because it is freely available, commonly used first, and directly connected to USPTO decision-making information.The comparison assumes paid services should outperform TESS on at least some measures.
  • TESS exploratory results: About half of TESS results returned a positive phonetic match under Soundex, while nonmatches remain important for identifying potential improvements.Soundex encodes consonants according to a predefined schema, but the authors note that it is simple and error-prone.
  • TESS exploratory results: Result counts balance exhaustive conflict coverage against the noise and human search costs created by too many results.The authors caution that the sampled distribution is only a baseline for what registrants can expect from TESS, not an inferential result.
  • Comparison framework: The study compares TESS and private engines using result counts, exact matches, phonetic matches, and close matches by letter substitution.The analysis treats these measures as complementary because trademark conflicts can involve spelling or sound.
  • Private-engine comparisons: Overall, AI-powered engines provide valuable applicant insights by adding information or presenting it more manageably, with different tools suited to different use cases.The results therefore suggest that engine choice depends on the applicant’s optimization objective rather than a single universal ranking.
  • Private-engine comparisons: Across 115 marks, Markify returned around 27,000 potential matches, compared with about 8,000 for TESS and Trademarkia and approximately 3,000 for TrademarkNow.The authors attribute this pattern partly to Markify’s use of additional sources and the other engines’ reliance on TESS.
  • Private-engine comparisons: Trademarkia and TrademarkNow returned similar numbers of exact matches to other engines while filtering more non-exact results, and their phonetic-match rates approached 40% and 50%, respectively.Corsearch returned fewer results overall but generally a higher proportion of phonetic matches, while private engines reduced noise relative to TESS.

7. Metrics

The metrics evaluate whether search engines find the marks underlying USPTO 2(d) rejections while balancing missed conflicts against irrelevant results. They combine classification measures to compare recall, precision, and related tradeoffs.

  • Evaluation target: The evaluation focuses on whether a search engine finds the “killer mark” supporting a USPTO 2(d) rejection.Finding the killer mark supplies information relevant to whether the proposed application will be accepted.
  • Classification measures: True positives are results matching a killer mark, false positives are non-killer results, and false negatives are killer marks absent from the results.True negatives cannot be identified because they would require both no returned results and no killer marks.
  • Classification measures: Recall is True Positive/(True Positive + False Negative), while precision is True Positive/(True Positives + False Positives).These measures capture different priorities: finding all relevant killer marks versus avoiding irrelevant results.
  • Evaluation design: The authors present the metrics for finding any killer mark, overall search results, and results limited per trademark application.This presentation examines engine performance under different result-volume conditions.

8. Results

Private trademark search engines provide meaningful value to potential registrants, although their performance differs across engines, metrics, and search-result limits. Compared with TESS, private engines generally improve recall, while engine-specific tradeoffs shape how many results applicants must review.

  • Private trademark search engines provide a genuine value-add to potential registrants, although meaningful differentiation exists among products.The analysis compares engines using returned results, match types, precision, and recall.
  • Most search engines fail to find a killer mark more often than not, and TESS performs comparatively well on finding at least one killer mark.A killer mark is sufficient to defeat a trademark application.
  • Every private search engine achieves higher recall than TESS without limiting returned results, and many also improve on precision.Precision and recall capture whether engines find relevant killer marks, including cases with multiple conflicts.
  • Search engines exhibit different precision-recall behavior as result limits change: Markify and TrademarkNow sustain roughly 0.55–0.60 recall, while Corsearch stabilizes near 0.40 and Trademarkia near 0.20.Markify and TrademarkNow may identify killer marks after about fifty results, whereas Corsearch and Trademarkia may require closer to one hundred.
  • The study concludes that AI-enabled trademark tools can efficiently compare proposed marks with registered marks and inform trademark-related decision-making.The broader implications extend to trademark search and legal administration.

B. FRAMING TRADEMARK REGISTRATION AS AN ADVERSARIAL MACHINE LEARNING PROBLEM

The paper frames trademark registration as an adversarial machine learning problem because applicants and the USPTO adapt their decisions in response to each other’s incentives and increasing sophistication. This reframing extends adversarial machine learning analysis to trademark search and examination.

  • Trademark registration may resemble an adversarial machine learning problem as applicants and the USPTO anticipate and adapt to each other’s incentives.Applicants seek to maximize claim scope, while the USPTO seeks to minimize it.
  • Adversarial machine learning describes applications in which underlying data distributions change in response to external stimuli.The paper applies this concept to the evolving interaction between trademark applicants and examination authorities.

C. RISK ASSESSMENT IN THE TRADEMARK ECOSYSTEM

AI-powered trademark tools can reduce search costs and support post-registration protection, but the paper treats trademark search as only the first step in broader brand management. It also identifies important limits and avenues for future research, including human oversight, broader data, and direct study of USPTO interactions.

  • AI-powered tools can substantially reduce search costs and bolster trademark holders’ ability to protect their intellectual property.They process large amounts of brand-related data, filter noise, and present information to brand owners.
  • AI should complement rather than replace human judgment because models trained on human decision-making data may reproduce human decision-makers’ limitations.The paper cautions against treating algorithmic judgment as fully independent from human judgment.
  • Trademark search is only the first step in overall brand management, whose AI-enabled services warrant further study.The same machine-learning tools power search and brand-management functions.
  • The study establishes a reproducible method that future researchers can extend with new searches, metrics, and data.Suggested extensions include pending applications, registrant types, and direct analysis of how AI search interacts with USPTO granting activity.
  • The paper concludes that AI makes trademark holders consumers of trademarks and shifts attention toward the supply of trademarks and legal administration.Its framework combines economic analysis with an empirical comparison of AI-related private search vendors.
Loading 2601.17072v1…